-
1
Objective corruption risk indicators using donor project and
contracts data
Mihály Fazekas* and Lili Márk**
*University of Cambridge and Government Transparency
Institute
**Central European University and Government Transparency
Institute
DRAFT, don’t cite without permission
Version 7/9/2017
1. Our approach to validating corruption risk indicators
To validate our red flags we use macro level perception-based
indicators and test if these
move together with the country-level averages of our red flags.
We also try to find testable
micro level relationships either on contract- or project-level
data and check if different red flags
are consistent with each other. The three donors largely differ
in the range of available
variables and red flags. World Bank data is the richest, while
we have much more limited
opportunities in case of IDB and EuropeAid.
As macro validation, we checked the correlations with some
well-established perception-based
corruption indicators on country-level (similarly to Fazekas and
Kocsis (2015)): World
Governance Indicators’ Control of Corruption, Transparency
International’s Corruption
Perception Index, and Global Competitiveness Index’s Favoritism
in decisions of government
officials (indicator 1.071). All three perception indices
indicate lower corruption with higher
values, so we expect to see negative correlations (Kaufmann,
Kraay, & Mastruzzi, 2009;
Transparency International, 2012; World Economic Forum, 2010).
This strategy has been
originally used for national procurement data and for
procurement notices published on Tender
Electronic Daily (TED), the procurement page of the European
Union; however, the corruption
risks of procurement from development aid sources might not go
hand in hand with the
corruption patterns of national procurement. Furthermore,
following from the regulations of the
donor institutions (Fazekas & Tóth, 2014) contracts below
country-specific thresholds are not
published on donor websites, thus we cannot even track the full
amount of development aid
1 In your country, to what extent do government officials show
favoritism to well-connected firms and individuals when deciding
upon policies and contracts? [1 = always show favoritism; 7 = never
show favoritism]
-
2
spent through corruption (see Figure 1-Figure 4 in the next
chapter). It might be the case that
suspicious transactions are managed below the threshold value
and larger contracts are kept
transparent. Consequently, we do not necessarily expect to see
strong correlations with these
indicators, but still, some level of correlation would
strengthen the validity of our red flags.
As micro validation, we explore the relationship between the
indicators of our own calculations
on contract-level data to see if they show a coherent picture.
For example, in order to use non-
open procedures as a red flag we should see positive correlation
with single bidding, showing
that non-open or restricted procedures indeed go together with
lower competition. We cannot
necessarily test whether there is a causal effect, but we mostly
check whether correlations
point to the expected direction and are significant. Our primary
approach is to estimate logit
models that predict single bidding using different set of
control variables, e.g. country, CPV
codes, sector, contract value. We treat single bidding as a
direct outcome of corrupt behaviour
and we would like to see how much the red flags that capture
suspicious aspects of the
procurement process correlate with single bidding. However, this
method works only if we have
a large enough sample size including several observations in
each country; otherwise we can
find spurious relationships. When single bidding is not
available we have limited possibilities
for micro validation, but we can always check the raw
correlations between separate indicators.
2. Share of aid spending captured by micro-level datasets
The figures in this chapter summarise how much of aid spending
we see in our databases
compared to the total aid spending of the donors. Figure 1 shows
the share of prior review
contracts in the case of World Bank. Prior review contracts have
to be published on the World
Bank website (red columns), whereas other contracts are only
published on the national
procurement websites (difference between light-green and red
columns). Thus, the black line
shows the share of lending amount for which contracts are
available to us in the World Bank
database. The publication rules are similar in case of the other
two donors.
Figure 1 Share of prior review contracts for World Bank
(1998-2013).
Source: World Bank, own calculations
-
3
Figure 2. Proportion of all DEVCO payments by management modes
(2014-2016).
Source: Annual Activity Report 2014, 2015, 2016
https://ec.europa.eu/info/publications/annual-activity-report-
2016-international-cooperation-and-development_en ,
https://ec.europa.eu/info/publications/annual-activity-report-
2015-international-cooperation-and-development_en ,
https://ec.europa.eu/info/publications/annual-activity-report-
2014-international-cooperation-and-development_en
Figure 2 shows the proportion of payments by the
Directorate-General for International
Cooperation and Development (DG DEVCO) by management mode.
Spending via direct and
indirect management refer to procurement spending via the
European Commission directly or
via the Beneficiary Countries or International Organisations and
Development Agencies
indirectly. According to the procurement guidelines2 each tender
in direct or indirect
management mode has to be published on the EuropeAid website, so
about half of the
development spending appears on EuropeAid website.
2
https://ec.europa.eu/info/publications/annual-activity-report-2014-international-cooperation-and-development_en
https://ec.europa.eu/info/publications/annual-activity-report-2016-international-cooperation-and-development_enhttps://ec.europa.eu/info/publications/annual-activity-report-2016-international-cooperation-and-development_enhttps://ec.europa.eu/info/publications/annual-activity-report-2015-international-cooperation-and-development_enhttps://ec.europa.eu/info/publications/annual-activity-report-2015-international-cooperation-and-development_enhttps://ec.europa.eu/info/publications/annual-activity-report-2014-international-cooperation-and-development_enhttps://ec.europa.eu/info/publications/annual-activity-report-2014-international-cooperation-and-development_enhttps://ec.europa.eu/info/publications/annual-activity-report-2014-international-cooperation-and-development_enhttps://ec.europa.eu/info/publications/annual-activity-report-2014-international-cooperation-and-development_en
-
4
Figure 3 Share of contract awards published on TED out of
tenders on EuropeAid
Source: Parsed EuropeAid website, parsed TED website
On Figure 3 we show how many of the tenders on EuropeAid we can
see also on Tender
Electronic Daily (TED). According to the procurement guidelines
only international calls for
tender have to be published on TED additionally to the EuropeAid
website. International calls
must be used when the contract value is above a certain
threshold (EUR 300,000 for supplies
and services and EUR 5,000,000 for works). To calculate this
share we used the EuropeAid
reference number to merge our parsed data from TED and EuropeAid
website to see how
many matches there are. The graph shows the share in numbers,
but not based on contract
value, but the graph can be taken as a lower bound for the share
of contract value published
on TED as the tenders we see on TED have higher values.
-
5
Figure 4 Share of contract amount in our IDB database out of
total IDB lending
Source: IDB parsed database, IDB Annual Reports 2014, 2015, 2016
http://www.iadb.org/en/about-us/annual-
reports,6293.html
On Figure 4 we depicted the share of contract value appearing in
our parsed IDB database out
of total new lending approved in a given year. Loans and
guarantees approved come from IDB
Annual Reports and show the amount approved in a given year. Sum
of contract amount is
from the data that we generated by parsing the IDB website and
adds up the value of contracts
signed in a given year.
3. Source-by-source results
World Bank
The correlations with perception-based indicators for our most
important red flag, single-
bidding, are presented in Table 1. Single bidding is our most
straight-forward red flag. In order
to secure that resources are allocated to specific favoured
individuals other competitors should
be somehow ruled out from competition. Unless fake competitors
are commissioned single
bidding is necessary, but not sufficient sign of a tender.
Unfortunately, we have data about the
number of bidders only until 2009 for World Bank; and we do not
have this data for the other
two donors at all.
http://www.iadb.org/en/about-us/annual-reports,6293.htmlhttp://www.iadb.org/en/about-us/annual-reports,6293.html
-
6
Table 1 Correlation of single bidding and perception-based
corruption indicators
TI - CPI (2009)
WGI - CoC (2009)
GCI - Fav (2009)
Weighted with number of contracts
Single bidding -0.20 -0.15 -0.20
1 / bidder nr.2 -0.19 -0.14 -0.19
Weighted with sum of contract values
Single bidding -0.18 -0.11 -0.15
1 / bidder nr.2 -0.17 -0.10 -0.14
Note: Only countries with more than 100 contracts are
considered
Figure 5 Relationship between single bidding ratio and WGI
Control of Corruption indicator.
Notes: The size of bubbles represents the number of contracts in
that country. Number of contract per country used as weights. Only
countries where there are more than 100 contracts are included.
Raw correlations are very close to zero and are insignificant,
but when dropping countries with
fewer contracts from the sample the correlations become higher
and more significant. In Table
1 we present correlation coefficients for countries with more
than 100 contracts and use total
value and number of contracts weights. We can see that all
correlation coefficients are negative
as we expected, but are not too high in absolute value. In
Figure 5, we depict the average
1998-2009 single bidder ratio with their 2009 WGI Control of
Corruption scores to illustrate the
relationship between the two. It is obvious that it is not a
very strong and well-defined
correlation, but it is evidently negative.
Correlation coefficients between our other red flags (average
for contracts before 2015) and
the perception indicators (2015) for the World Bank data are
summarized in the Appendix. As
we have expected, these relationships are not especially strong.
The strongest red flags seem
to be supplier tax haven registry and the indicator for project
cost overrun above 120%.
-
7
In the case of World Bank, we also use single bidding for
checking micro-level validity, by
checking to what extent our different red flags predict
single-bidding in a tender (Table 2).
Results are mostly in line with our expectations, except
supplier tax haven, where we see a
negative coefficient.
Table 2 Micro validity testing. Coefficients from logit models
predicting single bidding on tender level.
Red flag Coefficients on single bidding
Procedure is restricted, single source or consultancy
2.16** 1.98**
Non-open procedure 2.67** 2.67**
Consultancy spending 1.75** 1.49**
Supplier is from tax haven -0.31** -0.36** -0.22**
Supplier is from tax haven and small state 0.02 -0.32**
-0.20**
Signature period 93 days 0.01** 0.11** -0.039*
Missing signature period 0.50** 0.66** 0.34**
Log(contract value)
x x
Sector
x x
Region
x x
Country
x x
Year
x x
Other red flags
x Note: ** - significant at 0.01 level, * - significant at 0.1
level
We used the database of Winters (2014) to validate our red flags
also on the project level.
Winters (2014) analyses the Implementation Completion Reports of
World Bank projects using
text mining techniques and searches for key words that let us
assume the project was
captured. We are testing whether our red flags show higher
corruption risks in case of captured
projects. Again, we do not expect very substantial significant
results for several reasons.
Winters (2014) have data only for 598 projects out of the
approximately 6000 projects for which
we have red flags. We did not find any supporting evidence that
our red flags are in line with
the capture variable of Winters, but we do not think it would
undermine the validity of these red
flags.
IDB
In case of IDB, we had fewer red flags to check and also much
more limited possibilities for
testing validity. Table 3 summarises the results of our macro
validity checks. All the correlations
support the validity of our red flags more or less, except for
the non-open procedures indicator.
In case of this red flag the correlations looked better when
using value weights or number of
contracts weights.
-
8
Table 3 Correlation of the country-average of our red flags
until 2015 and perception-based corruption indicators of 2015 for
IDB
GCI - Fav WGI - CoC TI - CPI
Submission
Non-open procedure 0.07 0.13 0.24
Consultancy spending -0.21 -0.37 -0.34
Outcome
Supplier is from tax haven -0.24 -0.38 -0.37
Supplier is from tax haven and small state
-0.18 -0.39 -0.37
Publication share of awarded contracts
0.24 0.32 0.31
Note: for all red flags we expect to see negative correlations,
except for the publication share of awarded contracts
where we predict that a higher share of published contracts
would suggest lower corruption levels.
As we do not have information about the number of bidders for
IDB, we just checked how the
different red flags correlate with each other. We used a
project-level database to check
correlations as the share of published contracts red flags is on
the project-level. In case of
contract-level indicators we calculated simple mean within a
project. The results were mostly
positive but not really strong. All red flags except the share
of published contract awards are
expected to have positive correlations with each other.
Table 4 Micro validation for IDB data. Correlation coefficients
on project-level data.
Share of published contract awards
Non-open procedure
Consultancy spending
Supplier is from tax haven
Supplier is from tax haven and small state
Share of published contract awards
1
Non-open procedure
-0.0042 1
Consultancy spending
-0.0629* -0.0272 1
Supplier is from tax haven
-0.0167 0.01 0.1083* 1
Supplier is from tax haven and small state
-0.0105 0.0226 0.0947* 0.8449* 1
Note: everywhere, except for the first column, we expect to see
positive correlations.
EuropeAid
For EuropeAid the only source from where we could get meaningful
structured information is
the TED website. However, only a subset of all EuropeAid
tendering is published on the TED
-
9
website and for shorter time period. For the years 2011-2017
there is 1806 contract awards
and 2230 contract notices on the TED website.
Due to the low sample size we only included those countries in
the macro validity checks where
there are at least 30 awards during the whole period – covering
58% of all the contracts – and
checked the correlation of red flags with the 2015 value of the
perception-based indicators.
Only the red flag indicating long decision periods has negative
correlation with all three
perception indicators. Non-open procedures have substantial
negative coefficients with the
WGI Control of Corruption and the TI Corruption Perception
Indicator, but as shown in Table
6, the micro validation shows unexpected results for this
variable. Besides the above
mentioned two, single bidding, no call for tender matched,
advertisement period below the
minimum and above the typical values are the red flags that show
sign of validity.
Table 5. Correlation of the country-average of our red flags and
perception-based corruption indicators of 2015 for EuropeAid
GCI - Fav
WGI - CoC
TI - CPI
Non-open procedure 0.03 -0.58 -0.47
No CFT matched -0.28 0.06 -0.04
Advertisement period 0-30 days -0.04 -0.53 -0.46
Advertisement period >41 days -0.16 0.27 0.51
Length of eligibility criteria (binary: above 1.075*CPV avg)
0.38 0.42 0.41
Decision period >146 days -0.16 -0.26 -0.28
Single bidding -0.40 0.17 -0.04 Note: countries in the sample
are Algeria, Argentina, Egypt, Ethiopia, Ghana, Lebanon, Mali,
Nicaragua, Tunisia,
Turkey
In the micro validity testing (Table 6) we run logit models to
predict single bidding. Except non-
open procedure and no CFT matched all other red flags have a
positive coefficient in our model
specifications with varying significance levels.
Table 6. Micro validity testing. Coefficients from logit models
predicting single bidding on tender level.
Red flag Coefficents on single bidding
Non-open procedure -1.410*** -1.065*** -1.033***
No CFT matched -0.196 -0.376* -0.227
Advertisement period 0-30 days 0.534 0.547 0.461
Advertisement period >39 days 1.670*** 0.890** 1.160***
Length of eligibility criteria (binary: above 1.025*CPV avg)
0.871*** 0.831*** 0.822***
Decision period >136 days 1.415*** 0.649** 0.614*
CPV codes
x x
Countries
x
-
10
4. Summary of validity test results and indicator selection
In Table 7 we summarised the results of the validity testing.
For each red flag at each donor we
considered a red flag to have high level of validity if at least
two correlation coefficients are
above 0.1 and two logit model coefficients are significant and
positive and robust to
specifications. One correlation above 0.1 and one significant
coefficient would be categorised
as moderate level of validity. Low validity level means that the
direction of coefficients are as
we expected but quite weak. No validity is the case when the
coefficients point to the opposite
direction than what we have expected.
-
11
Table 7 Summary table of validity testing
World Bank
Indicator Definition Level Time span
macro validity
micro validity
shortlist
Single-bidding 1=1 bidder per contract 0=2 or more bidders per
contract
Tender (Contracts)
1998-2008
High x
Non-open procedures
1=non-open procedure types (e.g. single source) 0=open procedure
types (e.g. international competitive bidding)
Tender (Contracts)
1998-2016
High High x
Spending on consultancy
1=consultancy procured 0=non-consultancy type product
purchased
Tender (Contracts)
1998-2016
Low (regional variability)
High x
Signature period < 14 days
Time between award date and contract signature date is shorter
than 14 days
Tender (Contracts)
1998-2013
High High x
Advertisement period length < 14 days
Time between publication and bidding deadline is shorter than 14
days
Tender (Call for Tenders)
2009-2016
Moderate N/A** x
Supplier tax haven registration
1=Foreign supplier registered in a tax haven 0=Foreign supplier
registered in non-tax haven (or Domestic supplier)
Tender (Contracts)
1998-2016
Moderate No x
Share of published contract awards
Sum of contract awards amount / total project cost
Project 1998-2016
Low Moderate
Cost overruns (WB part)
Final project cost compared / original committed amount
Project 1998-2016
High High x
Inter-American Development Bank
Indicator Definition Level Time span
macro validity
micro validity
shortlist
-
12
Non-open procedures
1=non-open procedure types (e.g. single source) 0=open procedure
types (e.g. internat.comp.bidding)
Tender (goods& works)
1991-2016
No Low
Spending on consultancy
1=consultancy procured 0=non-consultancy type product
purchased
Tender 1991-2016
High Moderate x
Supplier tax haven registration
1=Foreign supplier registered in a tax haven 0=Foreign supplier
registered in non-tax haven (or Domestic supplier)
Tender 1991-2016
High Moderate x
Share of published contract awards
sum of contract awards amount / total project cost
Project 1991-2016
Moderate Low - Moderate
x
EuropeAid
Indicator Definition Level Time span
macro validity****
micro validity
shortlist
Single-bidding 1=1 bidder per contract 0=2 or more bidders per
contract
Tender (Contracts)
2011-2017 Moderate x
Non-open procedures 1=non-open procedure types (mostly
restricted) 0=open procedure type
Tender (Contracts or Call for tenders)
2011-2017 High No
No call for tender published
1=we can find the call for tender published on TED 0=we cannot
find the call for tender published on TED
Tender 2011-2017 Moderate No
Advertisement period 1=0-30 days*** / 1=above 39 days
0=other
Tender 2011-2017 High /
Moderate Moderate
/ High x
Decision period 1=
-
13
Notes: * Micro validity checks mean relationship with
single-bidding in case of World Bank and correlations across red
flags, especially with tax haven in case of IDB. ** Cannot be
linked to single-bidding, only macro validation is available *** 30
days is the minimum according to regulations **** As there are lot
of countries with only 1-2 contracts, we checked macro validity on
countries that have at least 30 contracts: High: for macro at least
two correlation coefficients with perception indicators are above
0.1; for micro at least two significant positive correlations in
logit models predicting single-bidding Moderate: for macro at least
one correlation coefficients with perception indicators are above
0.1; for micro at least one significant positive correlations in
logit models predicting single-bidding or consistent positive (not
necessarily significant) coefficients Low: very week relationship
No: results are of opposite direction compared to what was
expected
-
14
References
Fazekas, M., & Kocsis, G. (2015). Uncovering High-Level
Corruption: Cross-National Corruption Proxies Using Government
Contracting Data (GTI Working Papers No. GTI-WP/2015:02). Budapest.
Retrieved from
http://www.govtransparency.eu/index.php/2015/11/15/uncovering-high-level-corruption-cross-national-corruption-proxies-using-government-contracting-data/
Fazekas, M., & Tóth, I. J. (2014). Indicators of high-level
corruption using World Bank public procurement data.
Kaufmann, D., Kraay, A., & Mastruzzi, M. (2009). Governance
Matters VIII. Washington, DC: World Bank.
Transparency International. (2012). Corruption Perceptions Index
2012. Berlin: Transparency International.
Winters, M. S. (2014). Targeting, accountability and capture in
development projects. International Studies Quarterly, 58(2),
393–404. http://doi.org/10.1111/isqu.12075
World Economic Forum. (2010). The Global Competitiveness Report
2010–2011. Geneva, Switzerland: World Economic Forum.
-
15
Appendix A
Table 8 Macro validation for World Bank data. Correlation
between perception indicators (2015) and red flags (averages before
2015) for World Bank
GCI - Fav WGI - CoC TI - CPI
Submission
Procedure is restricted, single source or consultancy -0.23
-0.03 -0.14
Non-open procedure -0.24 -0.03 -0.27
Consultancy spending -0.07 -0.01 0.06
Advertisement period is
-
16
Appendix B Compiling datasets
World Bank
In case of World Bank we parsed or downloaded data from four
online sources on the World
Bank website. Additionally, we used the internal dataset of
World Bank including some more
variables than the online sources. For our final analysis we
used the World Bank database to
which we have added project-level information from the Bank’s
project details page. We
summarize the main information about each data sources in Table
9. In the first section of
the table the primary data sources are listed. In the second
section we describe the merged
datasets we generated from the data sources. In the second
section we included a column to
show how many observations we could match between the given data
sources.
Table 9 Summary of data sources for World Bank
Source Years Number of observations
Level of observation and connectivity
Data sources
Major contract awards3 2000- 131,860 -
Contract-level with project ID and WB contract number
World Bank Projects & Operations4 1947- 16,000 -
Project-level with project ID
Notices (WB website)5 2005-* 36,917 -
Contract-level with project ID and WB notice number
Contracts (WB website)6 2002-** 142,533 -
Contract-level with project ID and WB contract number
Internal World Bank Database
1998-2014
245,126 - Contract-level with project ID, no contract ID
Merged datasets
Years Number of observations
No. of matches
ID for merging
Parsed contracts + major contract awards 2000- 150,460
123,933
Contract-level, WB contract number
Parsed & major contracts + notices (tender-level data)
2002-** 185,283 2,499
Contract-level, WB contract number
Project details + contracts & notices, project-level 1947-
16,810 5,302 Project-level, project ID
Parsed & major contracts + notices (tender-level data) +
project details, tender-level
2002-** 185,283 181,670 Contract-level, project ID
WB Internal database + project details, project-level
1998-2014
17,239 6,028 Project-level, project ID
*There is a jump in the number of observations in 2009
**In 2002 there are only 1,723 contracts vs. yearly 10-13 years
in later years
3
https://finances.worldbank.org/Procurement/Major-Contract-Awards/kdui-wcs3/data
4 http://projects.worldbank.org/ 5
http://projects.worldbank.org/procurement/procurementsearch?lang=en&srce=both
6 See footnote 5
https://finances.worldbank.org/Procurement/Major-Contract-Awards/kdui-wcs3/datahttp://projects.worldbank.org/http://projects.worldbank.org/procurement/procurementsearch?lang=en&srce=both
-
17
EuropeAid
Our main data source for red flag generation and validation was
TED7 due to the higher
number of potential red flags we could generate compared to the
EuropeAid website.
Table 10. Number of observations at different stages of dta
cleaning and merging
Stage 1 2 3
Publication form type Original Freq.
Lots after cleaning lot number
Lots without CN duplicates
Contract Award (CA) 1,806 1,911 1,911
Contract Notice (CN) 3,260 3,231** 2,970
Prior Information Notice 2,596 - -
Total 7,662 5,142 4,881
**It is smaller than original data, because tender ID is missing
in 31 cases in the original data
In our original parsed dataset we had 1,806 contract awards and
3,260 contract notices (call
for tenders in TED terminology). To create our final dataset we
matched contract awards to
contract notices on the lot level. A contract notice often
includes several lots and sometimes
contract awards are published separately for the different lots.
A one-to-many merging could
be used in this case (i.e. for one CN more CAs can be merged),
but sometimes contracting
authorities do not find a suitable applicant right away and
relaunch a contract notice for some
lots later in new contract notices. So we parsed lot number to a
separate variable from the lot
titles in order to be able to match based on tender ID and lot
number. For the tender IDs that
match between CNs and CAs, we cleared the lot number
manually.
Identifiers used for matching:
Tender ID (e.g. EuropeAid/130735/D/SER/CO) is in fact the
EuropeAid reference
number, an ID for tenders that can be used to merge contract
notices to contract
awards.
Document ID (e.g. 2012/S 153-254700): each document has its
separate unique ID.
Lot title: the title of the separate lots in a tender, we can
gain the lot number from this
text, so that we can match based on tender ID and lot
number.
Steps
1. Original data
a. CN: tender-lot level,
b. CA: notice level, sometimes more lots in one CA notice
2. Generate cleaned lot-level data by parsing lot number from
lot titles
7 http://ted.europa.eu/TED/search/search.do Selecting “External
aid programmes” and “European Development Fund and External aid”
from the dropdown menu of European Institutions in the advanced
search.
http://ted.europa.eu/TED/search/search.do
-
18
3. Generate CN data with unique tenders: when there is a CN
relaunch for a lot, we
keep the latest version of the notice
Table 11. Result of matching CN and CA data
Result N
Not matched 2,319
Only CN 1,693
Only CA 626
Matched 1,285
IDB
We had three data sources for IDB: contract awards, procurement
notices and project
details8. We could not merge contract awards to call for tenders
due to the lack of a unique
ID. The only potential red flag from the procurement notice
database without merging it with
the contracts would have been advertisement period, but it
turned out that this variable
cannot be generated from the data. So we did not use the call
for tender data later on.
We generated our red flags separately for the contracts and
projects data sources and
merged them on the project level, where we can see the average
value of tender-level red
flags for a given project. We used this project-level data for
macro level validation and the
two datasets separately for micro validation. Table 12 show the
number of observations in
each data set before and after merging.
Table 12. Preparing IDB database
Years N Level of observation and
connectivity
Data sources
IDB Proc. notices 1999-* 15,441 Tender-level with project ID
IDB Contract awards
1961- 357,932
Tender-level with contract reference number and operation
number
IDB Project details 1960- 20,905 Project level with project
ID
Merged data
CA data collapsed to project level 1960- 4,232
Project level with project ID
CA data matched to project data 1961- 4,232
Project level with project ID
*Peak on Jan 1 2001 (28%), proper number of observations since
2012
8 Scraped from this website:
http://www.iadb.org/en/projects/project-procurement,8148.html
http://www.iadb.org/en/projects/project-procurement,8148.html